Pyomnet-AI: Enabling Efficient OMNeT++ Network Optimization Through Low-Overhead, Hybrid Python AI Integration

Integrating AI/ML workflows into OMNeT++ is increasingly important for adaptive network optimization, especially in 5G-Advanced and emerging 6G RAN studies, where learned policies can improve scheduling, mobility management, and resource utilization. However, pure C++ integrations are often insufficient because modern AI development is largely centered around Python ecosystems and toolchains. Additionally, existing Python-based AI integration approaches mostly rely on socket communication, introducing serialization and coordination overhead which is not well suited to latency sensitive or high call frequency integration scenarios. This paper therefore presents PyOMNeT-AI, a framework for AI integration in OMNeT++ built around three complementary patterns. PyCoSim enables bidirectional co-simulation with Python modules as first-class simulation components. PyEmbed provides scoped interpreter embedding for inference-centric workloads, achieving a baseline overhead of 0.28 μs and sustaining over 3 million calls per second. PyShm provides shared-memory coupling for Gymnasiumcompatible RL training, yielding a 2-6 × speedup over socketbased alternatives depending on workload and payload size. We present the design rationale, performance characterization, and pattern-selection guidelines, validated through Simu5Gbased RAN studies and designed to transfer to other OMNeT++ application domains.